Mobile Phone Diabetes Project Led To Improved Glycemic Control And Net Savings For Chicago Plan Participants

Abstract

Even with the best health care available, patients with chronic illnesses typically spend no more than a few hours a year in a health care setting, while their outcomes are largely determined by their activities during the remaining 5,000 waking hours of the year. As a widely available, low-cost technology, mobile phones are a promising tool to use in engaging patients in behavior change and facilitating self-care between visits. We examined the impact of a six-month mobile health (mHealth) demonstration project among adults with diabetes who belonged to an academic medical center’s employee health plan. In addition to pre-post improvements in glycemic control (p = 0.01) and patients’ satisfaction with overall care (p = 0.04), we observed a net cost savings of 8.8 percent. Those early results suggest that mHealth programs can support health care organizations’ pursuit of the triple aim of improving patients’ experiences with care, improving population health, and reducing the per capita cost of health care.

Effective treatments are available for many chronic diseases. However, such diseases are often poorly controlled, and they remain a leading cause of preventable morbidity, mortality, and excess costs.^1,2 A key challenge remains the health care delivery system, which is organized around doctors and hospitals instead of patients and communities.

Even with the best health care available, patients with chronic illnesses typically spend no more than a few hours a year in a health care setting, where most health care resources are directed. Outcomes for patients with chronic illnesses are largely determined by the activities they engage in during the remaining 5,000 waking hours of the year, for which fewer health care resources are available. These activities include taking medication, eating healthy food, engaging in physical activity, and monitoring symptoms.

Current approaches to supporting patients beyond the clinical encounter, such as remote care management, are personnel-intensive³ and have produced only modest or mixed results in effectiveness trials.^4,5 Classes that teach self-care can be effective for a range of chronic conditions.⁶ However, their effects are poorly sustained and limited by a lack of integration into clinical care.⁷ And home telemonitoring, in which patients transmit data on their health status using electronic devices, has failed to live up to expectations,⁸ in part because of poor patient engagement.⁹

Mobile phones are a promising platform for engaging patients in chronic care because most patients own and regularly use such a phone.¹⁰ Through text messaging and mobile applications, patients can receive at any time and any place health information that is targeted to them and is delivered instantly; they can also share information with their providers.

Several recent studies have successfully piloted programs based on mobile phone text messages, targeting patients with asthma, obesity, smoking, and diabetes.^11,12 In addition, a recent clinical trial of a mobile health (mHealth) diabetes coaching program demonstrated improved glycemic control among patients with difficult-to-control diabetes.¹³

These results are promising. However, evidence about the value of mHealth chronic care programs outside of research settings is limited. And the following questions remain unanswered: how to make these technologies engaging to diverse patient populations, how to integrate the technologies into clinical care, and how to automate programs sufficiently to reduce their personnel costs.

At University of Chicago Medicine, an academic medical center on Chicago’s South Side, we have pioneered a disease management program that is based on the use of mobile phones for members of the employee health plan. The program uses automated text-messaging software to engage patients in self-care and to facilitate care coordination between nurses at the health plan and physicians at the medical center.

We present the results of a quasi-experimental study that we conducted between May 2012 and February 2013 among participant and nonparticipant adults with diabetes, including clinical outcomes, utilization, and costs. For this study we chose to focus on diabetes because it is one of the best examples of a condition in which self-care behavior is linked to clinical outcomes.

Program Intervention

CareSmarts is an mHealth diabetes program that provides self-management support and team-based care management for people with diabetes through automated text messages. The program is a theory-driven behavioral intervention designed to improve self-care through multiple mediators, including cuing, education, self-efficacy, social support, and health beliefs.^14,15

Health plan nurses enroll eligible members in CareSmarts over the phone. Nurses use a web-based enrollment form that includes the patient’s mobile phone number, diabetes care plan, and preferred times for receiving messages.

Participants in the program receive text messages about diabetes self-care, some prompts to engage in a particular action (for example, “Time to check your blood sugar”), and some questions (such as, “Do you need refills of any of your medications?”) to which the patient responds by texting. The contents of the messages that the participants receive are modified through software every two weeks as needed, based on their interactions with the system.

Participants follow a flexible education curriculum in which they move from one topic to the next at their own pace. Each two-week education module focuses on one diabetes self-management topic (such as medications, nutrition, glucose monitoring, foot care, and exercise) and one topic related to living with chronic illness (such as navigating the health care system and managing stress).¹⁶

Once participants complete the curriculum, which lasts a minimum of ten weeks, they have the option of receiving further modules on nutrition and exercise, or stopping the education messages and receiving only reminders and self-assessments. Reminders about medication, glucose monitoring, and foot care are sent to participants at a frequency determined by participants’ preferences and responses to self-assessments, which are used to track their self-care behavior and determine when they need reminders.

Participants’ responses to self-assessment questions also facilitate remote monitoring and care management by health plan nurses. A response that is outside established parameters triggers an alert. Using protocols, nurses respond to each alert by the next business day.

“Low-level alerts” refer to care coordination issues, such as medication refills or referrals to diabetes educators. Following a low-level alert, nurses coordinate with the primary care team to process the refill or referral and then notify the patient that the issue has been resolved. “High-level alerts” refer to more serious problems, such as low self-reported medication adherence or poor patient responsiveness to questions. After a high-level alert, a nurse conducts a structured assessment of the patient by telephone and communicates the results to the primary care team by e-mail.

Nurses are not expected to make any scheduled outbound calls. They contact patients by telephone only in the case of exceptions—alerts—and thus provide what is called exception-based care management.

Study Data And Methods

STUDY DESIGN

The research design was a quasi-experimental, two-group, pre-post study. For reasons of fairness, the University of Chicago’s employee health plan required that every eligible member age eighteen or older with type 1 or type 2 diabetes be contacted for recruitment. For the purposes of this study, people enrolled in CareSmarts made up the treatment group; participants in the employee health plan not enrolled in CareSmarts served as the control group.

Data were analyzed on an intent-to-treat basis, meaning that patients who were enrolled in Care-Smarts but subsequently disenrolled were analyzed in the treatment group, using measures defined in advance. The benefit of this design was twofold: It accounted for secular trends, including ongoing quality improvement initiatives, and it allowed cost savings to be estimated on a population basis. The study was approved by the University of Chicago Institutional Review Board.

STUDY RECRUITMENT

The study population included all adult health plan members with a diagnosis of type 1 or type 2 diabetes (International Classification of Diseases, Ninth Revision [ICD-9], diagnosis codes 250.xx) receiving care at the University of Chicago Medicine (N = 348). Eligible members received a mailing and up to two recruitment phone calls inviting them to participate in the program. Those with poor diabetes control (hemoglobin A1c of 8 percent or more) received up to eight additional calls.

People without access to a personal mobile phone were excluded. Participants were responsible for any text messaging costs charged by their phone carriers. A $25 cash incentive was given to each participant at the completion of the study.

OUTCOME MEASURES

For the treatment group, outcomes were assessed using the patients’ electronic health records (EHRs), billing data, pharmacy claims, telephone-based surveys, and a database containing information about the text messages sent and received during the study. For the control group, EHR and billing data alone were available.

PATIENT SATISFACTION AND ENGAGEMENT

After six months, participants in the treatment group were asked about their satisfaction with CareSmarts using a six-point Likert response scale (1 was “strongly agree,” 6 was “strongly disagree”). In addition, CareSmarts provided reports including the overall number of text messages sent and received, response rates, and response times.

BEHAVIORAL DATA

The Summary of Diabetes Self-Care Activities Measure created by Deborah Toobert and colleagues¹⁷ was used to assess the five following areas of self-care practice on a one-week scale (with response options ranging from zero to seven days): healthy eating, fruit and vegetable consumption, exercising, blood sugar testing, and foot care. This is a brief, validated measure of self-management that correlates with other measures of physical activity and dietary patterns.

CLINICAL DATA

We obtained clinical data on HbA1c, lipid profile, body mass index, and blood pressure for members of both groups from the participants’ EHRs. For the treatment group, the Brief Diabetes Distress Screening Instrument¹⁸ was also used as a validated measure of diabetes-related quality of life. The instrument asks respondents to rate the degree to which the following items caused distress: “feeling overwhelmed by the demands of living with diabetes” and “feeling that I am often failing with my diabetes regimen.” Respondents use a six-point Likert response scale (1 was “not a problem,” 6 was “serious problem”) in their ratings.

QUALITY OF CARE

A single survey question asked participants in the treatment group, “How satisfied are you with the overall care you receive at your health plan?” They responded on a five-point Likert response scale (1 was “very satisfied,” 5 was “very unsatisfied”).

UTILIZATION AND COSTS

Data on participants’ utilization and costs were obtained from the medical center’s billing data, which included outpatient visits, emergency department visits and hospital admissions, and laboratory and radiology services. Utilization outside of University of Chicago Medicine, which typically accounts for less than 7 percent of the number of claims paid, was not included.

Pharmacy claims data were used to assess prescription drug costs for all medications and to calculate the proportion of days covered for all regularly administered medications for diabetes.

STATISTICAL ANALYSES

We conducted descriptive analyses of participants’ sociodemographic and clinical characteristics, and we used t tests to evaluate differences between treatment and control groups. Using paired t tests, we tested for the following pre-post differences between the groups: patient-reported measures, clinical outcomes, utilization, and costs.

To increase the statistical accuracy of our results, we measured and compared utilization and costs per day for the twelve months prior to enrollment in the program—the pre period—to those in the six months during enrollment—the post period. When presenting our results, we report findings on a six-month period for both the pre and post periods, for ease of interpretation. The statistical software SAS, version 9.3, was used for all analyses. The level of significance was 0.05.

PROGRAM COSTS

Variable costs were software, telecommunication, and nurse staffing. The CareSmarts technology was contracted from a software vendor, mHealth Solutions, which charged a per member per month rate inclusive of text messaging fees. Nurse staffing was tracked using weekly log sheets. Fixed costs were the medical director’s salary and benefits (for five hours per month) and a one-time software customization fee.

LIMITATIONS

The primary limitations of this study were its lack of randomization and its quasi-experimental design. Participants in the treatment and control groups had similar demographic characteristics and baseline glycemic control. However, participants in the control group may have differed from those in the treatment group because they did not respond to the recruitment mailing or phone calls; lacked a personal mobile phone or texting capability, and therefore were excluded; or refused to participate. Thus, our results might be subject to selection bias.

Additional limitations include the study’s brief duration and lack of long-term follow up; and incomplete data on the control group, including no pharmacy and telephonic survey data.

Study Results

PROGRAM ENROLLMENT

Of the 348 eligible adult health plan members with diabetes, 74 were initially enrolled in CareSmarts. Of the 257 members who were reached by phone or who returned our recruitment mailing, 61 (24 percent) were excluded because they lacked a mobile phone or texting capability. Another 94 members (37 percent) refused to participate, with the most common reasons being “unfamiliar or uninterested in texting” (n = 43), “diabetes already well-managed” (n = 14), and “too busy” (n = 9).

Of the seventy-four people enrolled in CareSmarts, sixty-seven completed the six-month intervention. The reasons for discontinuing were not finding the program helpful (n = 3), loss to follow-up (n = 3), and termination of health plan membership (n = 1).

STUDY PARTICIPANTS

The average age of participants in the treatment group was fifty-three years (range: 22–69 years), and 65.6 percent of that group were African American (Exhibit 1). The average duration of diabetes was eight years (data not shown). Of the participants in the treatment group, about one third each had well-controlled diabetes (HbA1c of 7 percent or less), moderately controlled diabetes (HbA1c greater than 7 percent but less than 8 percent), and poorly controlled diabetes (HbA1c of 8 percent or more).

EXHIBIT 1.

Characteristics Of 348 Adult Members Of University of Chicago’s Health Plan With Diabetes, May 2012–February 2013

Characteristic	Treatment group (n=74)		Control group (n=274)		p value
	Years	SD	Years	SD
Mean age	52.8	(9.2)	54.5	(11.2)	0.21a
Characteristic	Number	Percent	Number	Percent	p value
SEX
Female	40	54.0	140	51.7	0.71b
Male	34	46.0	131	48.3
RACE
Black	36	65.5	127	57.7	0.55c
White	14	25.5	70	31.8
Asian or Pacific Islander	4	7.3	21	9.6
Other	1	1.8	2	0.9
ETHNICITY
Hispanic	2	3.4	16	7.7	0.54c
Non-Hispanic	51	96.2	193	92.3
Characteristic	Percent	SD	Percent	SD	p value
Mean baseline HbA1c	7.9	(2.1)	7.5	(1.7)	0.08a

SOURCE Authors’ analysis of billing and electronic health record data. NOTES Numbers may not sum to totals because of missing data. SD is standard deviation. HbA1c is hemoglobin A1c.

^aStudent t test, pooled, equal variance assumption.

^bChi-square test.

^cFisher’s exact test, nondirectional, because one or more of cells has an expected frequency of 5 or less.

No significant differences in the demographic characteristics or baseline HbA1c control were observed between the treatment and the control groups.

PATIENT ENGAGEMENT

Participants sent and received an average of 3.4 text messages per day (range: 2.1–6.5 messages). On average, participants responded to 52 percent of the self-assessment questions (range: 0–91 percent), and this response rate remained relatively constant throughout the study period (the rate was 55 percent in month 1, 50 percent in month 2, 52 percent in month 3, 51 percent in month 4, 48 percent in month 5, and 52 percent in month 6). The average median response time was 19 minutes (range: 1–192 minutes).

PATIENT EXPERIENCE

Seventy-three percent of the participants in the treatment group were satisfied with the program, and 77 percent said that they would be willing to participate in a similar program in the future. Participants agreed that the text messages helped them with self-care. However, agreement varied by self-care activity, ranging from 59 percent for medication adherence to 77 percent for foot care.

Most participants agreed that phone calls from the nurse were helpful for education (64percent) and health care navigation (70 percent). Participants also reported that knowing a health professional was reviewing their messages was important for their engagement (88 percent).

CARE MANAGEMENT

The average participant triggered 6.1 nurse alerts over the six-month period (median 8 alerts; range: 0–19), or approximately 1 alert per month. Nearly two-thirds of the alerts were low-level alerts; about one-third were high-level alerts. As noted above, low-level alerts involve issues such as medication refills and referrals, while high-level alerts involve more serious issues such as low self-reported adherence to medications.

During the six-month study the average participant received six phone calls from a nurse, and providers received an average of two communications per participant from a nurse.

BEHAVIORAL RESULTS

CareSmarts participants’ self-care improved during the study period. The number of days in a seven-day period that participants reported following a healthy eating plan increased from 4.5 days to 5.2 days (p = 0.03), the number of days they reported monitoring their blood glucose rose from 4.3 days to 4.9 days (p = 0.03), and the number of days they reported practicing foot care increased from 3.6 days to 4.3 days (p = 0.01). Adherence to diabetes medications as measured by the proportion of days covered increased from 83 percent to 91 percent (p = 0.003).

CLINICAL RESULTS

Control of HbA1c improved in the treatment group: In the pre period HbA1c averaged 7.9 percent, and in the post period it averaged 7.2 percent (p = 0.01). Glycemic control also improved in the subset with poorly controlled diabetes: The average in the pre period was HbA1c of 10.3 percent, and in the post period it was 8.5 percent (p = 0.01); see Exhibit 2. No change in HbA1c was observed in the control group, including those with poorly controlled diabetes (data not shown).

EXHIBIT 2.

Clinical Outcomes, Costs, And Health Care Use Per Participant Per Six Months, In The Pre And Post Periods

	Pre period	Post period	Difference	p value
CLINICAL OUTCOMES
HbA1c	7.9%	7.2%	−0.4%	0.01
Diabetes-specific quality of lifea	2.6	2.0	−0.6	0.01
COSTSb
Outpatient	$2,624	$1,754	−$ 893	0.02
ED	70	16	−54	0.13
Inpatient	460	27	−439	0.11
Totalc	3,084	1,780	−1,332	0.004
NUMBER OF:
Outpatient visits	6.37	5.04	−1.35	0.01
ED visits	0.068	0.015	−0.047	0.11
Inpatient admissions	0.047	0.015	−0.033	0.07
Inpatient days	0.101	0.030	−0.073	0.11

SOURCE Authors’ analysis of billing and electronic health record data. NOTES There were seventy-four participants in the pre period and in the post period, except for HbA1c (sixty-one participants in the post period) and diabetes-specific quality of life (seventy-one participants in the pre period and sixty-four in the post period). Not all “difference” values may appear to be correct because of rounding or difference in observations. HbA1c is hemoglobin A1c. ED is emergency department.

^aLower scores on the Brief Diabetes Distress Screening Instrument (see Note 17 in text) indicate higher quality of life. A score of 3 is the threshold for diabetes distress.

^bOn January 31, 2013, University of Chicago Medicine changed billing systems. Therefore, cost data for participants whose six-month enrollment period extended beyond this date were imputed from earlier records (<0.3 percent of days observed). In addition, we imputed costs and numbers of visits, admissions, and days for people who did not have continuous plan membership during the observation period (n = 20; controls only).

^cIncludes outpatient, ED, and inpatient costs but excludes prescription drug costs.

In addition, no pre-post changes were observed in blood pressure, lipids, or body mass index in either the treatment or the control group. In the treatment group, quality of life improved (Exhibit 2).

QUALITY OF CARE

We measured patients’ satisfaction with the overall care provided by the health plan only in the treatment group. There, it improved from baseline (1.6 out of 5) to the end of the study (1.3 out of 5; p = 0.04).

UTILIZATION AND COSTS

In the treatment group, the number and costs of outpatient visits declined (Exhibit 2). Emergency department and hospital use and costs trended downward, but those changes did not reach statistical significance.

We also found that the number of diabetes-related visits—that is, primary care, endocrine, and ophthalmology visits—per patient was un-changed, while the number of visits to non–diabetes specialists decreased (3.5 visits per patient per six months in the pre period and 2.2 visits in the post period; p = 0.007).

The total cost of health care declined by $812 per participant per six months. This reflected a $1,332 decline in costs for outpatient, emergency department, and inpatient visits (Exhibit 2), which was partially offset by an increase in prescription drug costs of $520 (p < 0.0001; data not shown).

Six-month program costs were estimated to be $375 per participant ($150 for technology and $225 for staff), of which $205 are fixed costs ($60 for technology and $145 for staff). This suggests a net cost savings of $437 per participant, or an 8.8 percent savings over pre-period costs.

In the control group, no differences were observed in the combined costs per six months of outpatient, emergency department, and inpatient visits between the pre period ($3,410) and the post period ($2,500; difference: −$910; p = 0.08). Cost data were not available for prescription drugs in the control group. However, even if we assume for the control group the same increase in prescription drug costs that we observed in the treatment group and include the nonsignificant cost reduction that we saw in the control group, the program would still produce net cost savings.

In addition, total costs over the entire study population (both treatment and control groups) decreased from the pre period to the post period (p = 0.02), despite our observing a significant decrease in total costs only for the treatment group.

Discussion

This study evaluated results from a demonstration project of CareSmarts, a mobile phone–based program that provides automated self-management support and facilitates team-based care for people with chronic illnesses. Among adults with diabetes enrolled in an academic medical center’s employee health plan, we found evidence of improved clinical outcomes, greater patient satisfaction, and lower health care utilization and related costs.

Our results suggest that the program was associated with a total net cost savings of $32,388 over six months, which was an 8.8 percent savings over pre-period costs. Only 20 percent of the study population participated in the study intervention. However, the decrease in costs for that treatment group drove an overall decrease in costs across both the treatment and control groups. This suggests that the program was cost saving at the population level.

The major innovation of this model is how it leverages mobile technology to enable existing health system resources to support chronic disease care. Importantly, the mobile phone program is not a stand-alone app. Instead, it is a highly interactive program that is integrated into the health care system. Because the program is largely automated, full-time staff members are not required: Part-time nurses are able to enroll patients and respond to alerts when exceptions to normal results or responses are noted.

Unlike conventional care management models, which typically have staffing ratios of 30–100 patients per full-time employee, our model allowed for 400 enrollees per full-time employee. Moreover, the model had significant buy-in from primary care physicians, who were able to coordinate care with nurses when serious clinical issues arose. This integration of the program into the primary care team may partly explain the observed decrease in outpatient visits among participants in the treatment group.

Our model has several advantages over existing care management programs and self-management interventions. First, the program is “high touch,” engaging patients daily. Second, because the program is delivered remotely, and largely through asynchronous communication, patients have a low burden of participation, and the program is accessible wherever a patient happens to be.

Finally, the program emphasizes self-management instead of clinical care. There are few clinically urgent alerts, and the role of the nurse can be filled by diabetes educators, medical assistants, or health coaches with minimal training.

Policy Implications

Our demonstration project had important successes. However, it is not clear whether CareSmarts will transition into a fully supported program and be scaled up to serve the broader patient population at University of Chicago Medicine. Next we discuss policy issues that might affect the widespread diffusion of mobile health programs and suggest actions that can address those issues.

ALIGNING FINANCIAL INCENTIVES

Support for the CareSmarts program grew out of the shared mission of a medical center and an employee health plan to provide high-quality care to employees and their dependents, as well as out of financial incentives for lowering costs that the two organizations shared. However, for patients in other health plans, the medical center is largely paid on a fee-for-service basis and would face negative financial implications if it participated in a program that reduced health care use among those patients.

The growing trend toward shared savings programs and accountable care organizations, in which payers and providers take responsibility for the health of a defined patient population and share cost savings, will be important for the sustainability and diffusion of mHealth programs. Providers and payers should not wait for broad national payment reforms to align financial incentives, however.

As our study illustrates, large health systems have employed populations for whom financial incentives are already aligned, and a business case exists for the use of mHealth. Health systems should begin implementing mHealth programs among their own employees to build local capacity and experience with mobile technologies.

Aligned financial incentives also exist where an insurer and health system can work together to define a program and population for shared savings. Insurers should reach out to their predominant care delivery partners (and vice versa) to support the use of mobile technology that is clinically integrated into the workflow and staffing of care delivery organizations.

INTEGRATION WITH ELECTRONIC HEALTH RECORDS

Although a significant level of clinical integration was achieved, efforts were limited by the fact that CareSmarts is not integrated with the health plan’s EHR system. If these systems were integrated, appointment information could be used to trigger appointment reminders, and patients’ responses to questions could be shared directly with primary care physicians.

Accelerating efforts to promote interoperability between EHR systems and other software applications would greatly increase the impact of mHealth programs. The federal government could also play a role in incentivizing the integration of mobile health technology into EHR systems through the proposed stage 3 meaningful-use criteria for patient-generated health data.¹⁹ The final criteria should include the ability to add to EHRs patient data from mHealth applications.

REGULATORY UNCERTAINTY

Mobile health applications raise new concerns about patient safety and privacy. Such concerns limited our program’s availability to broad patient populations because we were required to get Institutional Review Board approval for our demonstration project—which required documentation of patient consent before enrollment. In September 2013 the Food and Drug Administration (FDA) released its final ruling on mobile medical applications.²⁰ The ruling indicates that mHealth applications such as CareSmarts will not routinely be subject to FDA regulation, although applications that interface with medical devices, such as glucometers and blood pressure monitors, will be regulated—a policy that could considerably hamper progress toward creating a connected health ecosystem.

The recently released FDA ruling did much to reduce regulatory uncertainty about mobile devices. However, uncertainty remains about the differences between the uses of devices for disease and those for wellness, the precise definition of an accessory to a medical device, and the use of apps that support medical decision making. In addition, more work is needed to clarify the overlapping roles of the FDA, the Office of the National Coordinator for Health Information Technology, and the Federal Communications Commission.

Finally, federal guidance is needed on whether and when text messaging can be considered secure in accordance with the rules of the Health Insurance Portability and Accountability Act (HIPAA) of 1996. The lack of clear guidelines and regulations concerning provider-to-patient communication in mHealth makes health care organizations less likely to develop innovative programs in this area.

The regulatory issues in mobile health are complex. Nonetheless, regulators can take multiple steps to reduce uncertainty and foster innovation, such as announcing specific timelines for when key issues will be clarified and identifying model mHealth programs and hypothetical cases that steer clear of high-risk regulatory issues.

Conclusion

University of Chicago Medicine’s experience with a mobile health diabetes program suggests that connected health solutions hold promise for supporting chronic disease self-care, improving clinical outcomes, and reducing costs. Our study offers early evidence that mHealth can enable health care organizations to effectively support patients beyond the traditional health care setting and achieve the triple aim of better health, better health care, and lower costs.

Although we found a business case for the use of mHealth, the diffusion and sustainability of mHealth depends on a supportive policy environment. Accelerated movement toward accountability for population health, increased interoperability with electronic health records, and clearer regulatory guidance will be important for unlocking mHealth’s potential to support behavior change and chronic care.